Overview
This project involved processing 4D motion capture data to drive a procedurally animated wheelchair asset, built for use in mobile AR experiences tied to the Paris Olympics coverage. The core challenge was inferring accurate wheel rotation and overall movement from per-frame volumetric capture meshes that had varying point counts and no consistent skeletal structure to read from directly.
Motion Estimation
The 4D capture provided a sequence of per-frame meshes representing the athlete and wheelchair together, but with no fixed topology between frames. A VEX-based solver was written to extract best-fit motion from the changing geometry, computing velocity and direction vectors frame by frame. These were fed into a CHOPS network to produce smooth, integrated rotation values for the wheels and a cleaned root animation for the chair body.
A KineFX rig was then built on top of this data, with wheel rotation driven procedurally from the extracted motion rather than manual keyframes. This meant the animation automatically adapted to any input capture without needing rework per shot.
Asset Optimisation for Mobile AR
Since all assets were destined for mobile AR deployment, polygon counts needed to be kept tightly controlled. The wheelchair model was built with topology efficiency as a primary constraint alongside visual quality, using normal maps baked from a higher resolution version to recover surface detail within the polygon budget.